Homework 1. 5.1. First notice that the sequence max n≤t≤n+1 B(t) − B(n) n = 1, 2, · · · is independent. By reflection principle, max n≤t≤n+1 d B(t) − B(n) = max B(n + t) − B(n) = max B(t) = |B(1)| 0≤t≤1 0≤t≤1 Consequently, for any θ > 0, o n o n p p P max B(t) − B(n) ≥ θ 2 log n = P |B(1)| ≥ θ 2 log n n≤t≤n+1 n o − θ2 +o(1) 2 (n → ∞) = exp − θ + o(1) log n = n Thus, ∞ n X P n=1 max n≤t≤n+1 By Borel-Cantelli lemma, o p B(t) − B(n) ≥ θ 2 log n B(t) − B(n) √ lim sup max 2 log n n→∞ n≤t≤n+1 ( ( =∞ θ<1 <∞ θ>1 ≥θ θ<1 ≤θ θ>1 a.s. Letting θ → 1 leads to the conclusion. 5.2. Write B(t) = B1 (t), · · · , Bd (t) , where B1 (t), · · · , Bd (t) are 1-dimensional independent Brownian motions. Clearly, |B(t)| ≥ B1 (t). Therefore, by the law of the iterated logarithm for 1-dimensional Brownian motion (Theorem 5.1), lim sup √ t→∞ |B(t)| B1 (t) ≥ lim sup √ =1 2t log log t 2t log log t t→∞ It remains to establish the upper bound. (b1 , · · · , bd ) in Rd , write a·b= d X j=1 For each a = (a1 , · · · , ad ) and b = aj bj and |b| = d nX j=1 Here we use the fact that |x| = sup a · x |a|=1 1 a.s. x ∈ Rd b2j o1/2 Given a ǫ > 0, we have {x ∈ Rd ; |x| = 1} ⊂ [ {x ∈ Rd ; a · x > 1 − ǫ} |a|=1 By finite cover theorem, there are integer m = m(ǫ) ≥ 1, and a1 , · · · , am ∈ Rd such that |ak | = 1 (k = 1, · · · , m) and d {x ∈ R ; |x| = 1} ⊂ m [ k=1 {x ∈ Rd ; ak · x > 1 − ǫ} For any x ∈ Rd with x 6= 0, therefore, x ak · ≥ 1−ǫ 1≤k≤m |x| max This implies that 1 max (ak · x) 1 − ǫ 1≤k≤m |x| ≤ In particular x ∈ Rd ak · B(t) |B(t)| 1 lim sup max √ ≤ 1 − ǫ t→∞ 1≤k≤m 2t log log t 2t log log t t→∞ 1 ak · B(t) 1 = max lim sup √ a.s. = 1 − ǫ 1≤k≤m 1−ǫ 2t log log t t→∞ lim sup √ where the last step follows Theorem 5.1 and the fact that for each k, ak · B(t) is an 1-dimensional Brownian motion. Finally, letting ǫ → 0+ on the right hand side leads to the requested upper bound. 5.11. Write n 1 1X 1{Sk >0} #{k ∈ {1, · · · , n}; Sk > 0} = n n k=1 and L{t ∈ [0, 1]; Sn∗ (t) > 0} = Z 1 1{Sn∗ (t)>0} dt = 0 n Z X k=1 k n k−1 n 1{S(k−1)+(nt−(k−1))Xk >0} dt For each 1 ≤ k ≤ n, we consider two possibilities: First, Sk−1 Sk > 0. In this case , nk ] and therefore the continuous curve S(k −1)+(nt−(k −1))Xk has no zero point on [ k−1 n 1{Sk >0} and n Z k n k−1 n 1{S(k−1)+(nt−(k−1))Xk >0} dt 2 are either 0 at same time, or 1 at same time. In the case Sk−1 Sk < 0, we use the bound Z 1{S >0} − n k k n k−1 n Summarizing our discuss, 1{S(k−1)+(nt−(k−1))Xk >0} dt ≤ 1 n 1X 1 1{Sk−1 Sk <0} L{t ∈ [0, 1]; Sn∗ (t) > 0} − #{k ∈ {1, · · · , n}; Sk > 0} ≤ n n k=1 Thus, all we need is to show n 1X P{Sk−1 Sk < 0} = 0 n→∞ n lim k=1 Indeed, given M > 0, {Sk−1 Sk < 0} ⊂ {|Xk | ≥ M } ∪ {|Sk−1 | ≤ M }. Thus n n k=1 k=1 1X 1X P{Sk−1 Sk < 0} ≤ P{|X| ≥ M } + P{|Sk−1 | ≤ M } n n Given ǫ > 0, √ lim sup P{|Sk−1 | ≤ M } ≤ lim P{|Sk−1 | ≤ ǫ n} = P{|U | ≤ ǫ} k→∞ k→∞ where N ∼ N (0, σ 2 ), σ 2 is the variance of the i.i.d. sequence, and the last step follows from the central limit theorem. Let ǫ → 0 on the right hand side: lim P{|Sk−1 | ≤ M } = 0 k→∞ This leads to n 1X P{|Sk−1 | ≤ M } = 0 n→∞ n lim k=1 Thus, n 1X P{Sk−1 Sk < 0} ≤ P{|X| ≥ M } lim sup n→∞ n k=1 Letting M → ∞ on the right hand side completes the proof. 5.15. (a) Write Tn = 1 min{j; |S(j)| = n} and τ = {t ≥ 0; |B(t)| = 1} n2 3 For any x > 0, write tn = [n2 x]. n o n P{Tn ≤ x} = P min{j; |S(j)| = n} ≤ tn = P max |S(j)| ≥ n j≤tn o By invariance principle, and a argument similar to Theorem 5.25, 1 d √ max |S(S(j)| −→ max |B(t)| t≤1 n j≤n Notice that n ∼ √ x−1 tn (n → ∞). We have n o n o n o lim P max |S(j)| ≥ n = P max |B(t)| ≥ x−1/2 = P max |B(t)| ≥ 1 = P{τ ≤ x} n→∞ j≤tn t≤1 t≤x Summarizing our argument, lim P{Tn ≤ x} = P{τ ≤ x} n→∞ x>0 (b) We first claim that ∞ p 1 X 1{ρ̃i−1 ≤n} − 1{ρ̃i−1 ,ρ̃i ≤n} −→ 0 n2 i=1 (1) Indeed, 1{ρ̃i−1 ≤n} − 1{ρ̃i−1 ,ρ̃i ≤n} = 1{ρ̃i−1 =n, and ρ̃i =n+1} P{ρ̃i−1 = n, ρ̃i = n + 1} = P{ρ̃i−1 = n}P{ρ̃i = n + 1|ρ̃i−1 = n} n+1 = P{ρ̃i−1 = n} 2n where the last step follows from (5.14). In addition, for any i ≥ n + 1 P{ρ̃i−1 = n} = X (x1 ,···,xi−1 ) P{ρ̃1 = x1 , · · · , ρ̃i−1 = xi−1 } where the summation is over all possible paths with x1 = 1 and xi−1 = n. By Lemma 5.31, P{ρ̃1 = x1 , · · · , ρ̃i−1 = xi−1 } = P{ρ̃0 = x1 }P{ρ̃1 = x2 , · · · , ρ̃i−1 = xi−2 |ρ̃0 = x1 } = n2−(i−2) 4 The number of the paths is at most i−2 n−1 In summary, ∞ X ∞ X i−2 2−(i−2) P{ρ̃i−1 = n, ρ̃i = n + 1} ≤ n n−1 i=n+1 i=n+1 ∞ X i−2 2−(i−n−1) = n2−(n−1) 2n = 2n = n2−(n−1) n−1 i=n+1 where the third step follows from Taylor expansion ∞ X 1 i−2 = xi−n−1 n−1 (1 − x)n i=n+1 |x| < 1 Hence, ∞ ∞ 1 X 1 X 2 ≤ 1 − 1 P{ρ̃i−1 = n, ρ̃i = n + 1} ≤ −→ 0 E {ρ̃i−1 ≤n} {ρ̃i−1 ,ρ̃i ≤n} 2 2 n n n i=1 i=n+1 So we have (1). We now claim that ∞ 1 X p (τi − τi−1 − 1)1{ρ̃i−1 ≤n} −→ 0 2 n i=1 Write Ln = max{j; ρ̃j = n} We have that ∞ X i=1 (τi − τi−1 − 1)1{ρ̃i−1 ≤n} = Ln X i=1 (τi − τi−1 − 1)1{ρ̃i−1 ≤n} Given ǫ > 0 and M > 0, Ln n X o 2 P (τi − τi−1 − 1)1{ρ̃i−1 ≤n} ≥ n ǫ i=1 2 ≤ P{Ln ≥ n M } + P n k X o (τi − τi−1 − 1)1{ρ̃i−1 ≤n} ≥ n2 ǫ max 2 k≤M n i=1 5 (2) Let S(j) be a simple random walk and write γn = min{j; S(j) = n}. By Lemma d 5.37, Ln = γn . Consequently, n o 2 2 P{Ln ≥ n M } = P{γn ≥ n M } = P max S(j) ≤ n 2 j≤n M n 1 o 1 o √ √ √ −→ P max B(t) ≤ =P max S(j) ≤ t≤1 M M n2 M j≤n2 M 1 n as n → ∞, where the last step follow from invariance principle. Notice that the right hand side can be made arbitrarily small by making M large. To establish (2), therefore, all we need is to show lim P n→∞ n k X o (τi − τi−1 − 1)1{ρ̃i−1 ≤n} ≥ n2 ǫ = 0 max 2 k≤M n (3) i=1 Let n be fixed for a moment and write Mk = k X i=1 (τi − τi−1 − 1)1{ρ̃i−1 ≤n} k = 1, 2, · · · We claim that Mk is a martingale with respect to the filtration {F + (τk )}. Indeed, h i + + E Mk |F (τk−1 ) = Mk−1 + E (τk − τk−1 − 1)1{ρ̃k−1 ≤n} F (τk−1 ) and h i i h + + E (τk − τk−1 − 1)1{ρ̃k−1 ≤n} F (τk−1 ) = 1{ρ̃k−1 ≤n} E (τk − τk−1 − 1)F (τk−1 ) = 0 where the last step follows from the proof of Lemma 5.32. By maximal inequality, P n o max 2 |Mk | ≥ ǫn2 ≤ k≤M n 1 ǫ2 n4 E max 2 |Mk |2 ≤ k≤M n 4 ǫ2 n4 2 2 E MM n2 2 Mn Mn i2 4 X h 4 X = 2 4 E (τi − τi−1 − 1)1{ρ̃i−1 ≤n} ≤ 2 4 E (τi − τi−1 − 1)2 ǫ n i=1 ǫ n i=1 ≤ 4 ǫ2 n4 M n2 C −→ 0 where the fifth step follows from Lemma 5.32. So (3) holds. By (1) and (2), the problem is reduced to the proof of Z ∞ ∞ 1 X d 1{|W (t)|≤1} dt (τi − τi−1 )1{ρ̃i−1 ≤n} −→ n2 i=1 0 6 (4) By definition, ρ̃i−1 = ρ(τi−1 ) and |ρ(t) − ρ(τi−1 )| ≤ 1 for τi−1 ≤ t ≤ τi . Thus, Z τi τi−1 1{ρ(t)≤n−1} dt ≤ (τi − τi−1 )1{ρ̃i−1 ≤n} ≤ Z τi 1{ρ(t)≤n+1} dt τi−1 i = 1, 2, · · · By the fact that ρ(t) = |W (t)|, 1 n2 Z 0 ∞ Z ∞ ∞ 1 1 X (τi − τi−1 )1{ρ̃i−1 ≤n} ≤ 2 1{|W (t)|≤n−1} dt ≤ 2 1{|W (t)|≤n+1} dt n i=1 n 0 By Brownian scaling, for n > 1, 1 n2 Z 0 ∞ (n ± 1)2 1{|W (t)|≤n±1} dt = n2 d This leads to (4). 7 Z 0 ∞ 1{|W (t)|≤1} dt